“More models trained with Direct Preference Optimization (DPO) will be released than any other type in the next 6 months.”